Enhancing Customer feedback processing with Machine Learning in Microsoft Azure

Abstract

Text Classification and Natural Language Processing (NLP) is developing fast, and all the applications are rapidly evolving, GPT-3 emerged in the field just last year and there are some open-source options in the field, like GPT-NEO or GPT-J. It is well known that data is the new oil. Especially the analytics side of NLP is yet to unravel. This thesis introduces a process and Microsoft Azure cloud architecture to preprocess and classify incoming text data, in this specific case, Finnish Tax Administration customer feedback. The thesis will go through all the steps needed to preprocess the text to discover named entities, use Named entity recognition (NER) for pseudonymization, use machine learning (ML) for classifying purposes, and determine the optimal ML model for organizing the customer feedback. The thesis divides into three main parts; the first part goes through the theoretical concepts of machine learning and state-of-the-art Natural Language Processing. The second part covers the essential components of the Microsoft Azure cloud and the options to use ML models and customer feedback processing. Also, the second section will cover the basic architecture for the application pipeline in Azure. The third part focuses on developing the model itself and testing different models to find the best performing classification model for the task at hand, the classification of the customer feedback data. We will be using SVM, Neural Networks, such as TensorFlow, and the coding is in Python.